Key Takeaways
- Every data point has more than one clock: when it was observed, when the source published it, when it was retrieved, and when it was used.
- Freshness is relative to cadence. A weekly report that is three days old is current; an intraday feed that is three hours old is not.
- A revision creates a new vintage. A point-in-time query returns only vintages that were available at the requested time, which keeps a backtest honest.
- A missing value stays missing. It is never filled with zero.
- Source tier is shown beside freshness and never merged into it.
- Live feeds are not connected. The tool works on the records you paste or on a clearly labelled illustrative sample.
Observed, Published, Retrieved and Used: What Is Each Date?
A single column called "date" hides four different facts. The inspector keeps them apart and labels each one.
- Observation date
- When the underlying condition was measured. A weekly futures-positioning report might describe positions held on a Tuesday.
- Publication date
- When the source made the value public. The same positioning report appears days after the Tuesday it describes. Until then nobody outside the source could act on it.
- Retrieval time
- When a system ingested the value. Retrieval cannot precede publication for a public source, so the inspector flags it if it does.
- Available time
- The earliest moment the value may enter a point-in-time analysis. It defaults to the publication time and is never earlier than it.
Dates that carry no clock time stay date-only, exactly as supplied. Converting a date to midnight in some time zone can move it to a neighbouring calendar day, so the inspector never does that. Date-times must include Z or a numeric offset, and every instant is displayed in UTC.
The guide Sentiment Data Latency, Publication Lags and Vintages explains why the gap between the first two dates matters so much for research.
Market Data Freshness Inspector
The sample is loaded for you. Press the button to inspect it, change the evaluation time, or replace the records with your own. Each record is a JSON object: seriesId, publishedAt and cadence are the essentials, and observedAt, retrievedAt, value, sourceTier, vintages and derivedFrom unlock the rest.
Illustrative sample data. The values, dates and series in the box below are fictional and are not current market data. Live feeds are not connected.
All processing runs in your browser. The records you paste are not sent to Swoopr Investment's servers, stored, or logged. The time zone for every instant shown is UTC.
How Are Freshness States Decided?
Each series is compared with the next release its own source schedule expects. The state is relative to that schedule, so a three-day-old value can be current while a three-hour-old one is stale. An unchanged value is never read as a failure: only a missed release moves a series past "pending".
| State | Meaning |
|---|---|
| Current | The next expected release is still comfortably ahead. |
| Update expected soon | The next release is due within a short, cadence-specific window. |
| Expected release pending | The release time has passed but the grace period for that cadence has not. |
| Stale | The grace period has also passed with no new release. |
| Source unavailable | The grace period has passed and you reported that the source request failed. |
| Schema error | A required timestamp is missing or unreadable, or you flagged a schema change. The last known value is kept and shown as degraded. |
| Revision detected | A new vintage arrived for a value that is otherwise on schedule. |
| Unknown cadence | No recognised cadence was given, so the age is shown but no schedule is judged. |
Every state is shown as words plus a symbol, never by colour alone. When one required input of a composite misses its release, the composite names the affected input instead of marking everything stale.
What Are Revisions, Vintages and Look-Ahead Bias?
Many macro and survey series are revised after their first release. A revision is not an overwrite: it is a new vintage, identified by the series, the observation period and the time that version became available. The inspector keeps every vintage and answers two different questions.
- Latest revised: the newest value known today.
- Point in time (as of): the newest vintage that was actually available at a chosen past timestamp. A revision published after that timestamp is excluded, however accurate it later turned out to be.
Look-ahead bias is the error of using information that was not yet public at the time being studied. Using a revised figure for a date before the revision, or a month-end statistic before its release day, makes a historical result look better informed than any real participant could have been. A missing value is reported as missing. It is never treated as zero, because zero is a real reading.
See the guide on data latency, publication lags and vintages for worked examples.
Which Inputs Sit Behind a Derived Series?
A composite value inherits the timing of everything that feeds it. For a series with a derivedFrom list, the inspector lists every contributing input with its version, availability time, cadence and state. Required inputs limit the derived series' freshness. Optional inputs are shown but do not hold it back, and an input that became available after the derived value was calculated is flagged.
The point is proportion. One monthly input that is within its own cadence does not make a daily dashboard stale, but a required input that has missed its release is named, so the affected evidence family is visible. The sentiment composite framework describes how required inputs are chosen.
Why Is Freshness Separate From Source Authority?
Freshness says how recently a source updated relative to its schedule. Authority says how much weight the source deserves. The two are independent: a current alternative-data series is not more authoritative than a primary report that is a day late. The inspector shows a source tier from 1 (primary or official) to 4 (alternative or unverified) in its own column, using the ideas in the source ladder, and never folds it into the state.
Two data families the tool is built around are described in Commitments of Traders (weekly, observed on Tuesday, published days later) and margin debt and leverage (monthly, published after month end).
About the Illustrative Sample
The bundled sample is fictional. It holds eight series with made-up identifiers and made-up dates around a fixed evaluation time of 2026-10-05 14:00 UTC: an intraday feed, a daily close, a weekly positioning report, a monthly margin statistic, a quarterly filing metric, a revised macro series with three vintages, a search-interest index and a derived composite. None of the numbers describe a real market or a real publication.
Live feeds are not connected. For real values, go to the primary source and paste its publication timestamps into the inspector. Weekly positioning reports come from the CFTC; the notes linked in the References explain what the report measures and when it is released.
What This Tool Does Not Do
- It fetches nothing. It judges the timestamps you provide and cannot confirm that they are correct.
- It does not know release calendars. It uses a generic cadence rule plus the holidays you list. A source with an irregular schedule can give an explicit next-release time.
- It does not rank data quality. A current state says the schedule is on track, not that the figure is accurate.
- It does not forecast. Nothing here is a signal, a prediction or investment advice.
Frequently Asked Questions
What is the difference between observation date and publication date?
The observation date is when the underlying condition was measured. The publication date is when the source made the value public. A weekly positioning report can describe a Tuesday and appear on a Friday, so for three days the Tuesday figure existed but nobody outside the source could use it. Any historical study has to line data up by publication date, not by observation date.
What is a data vintage?
A vintage is one specific version of a value as it stood at a point in time. Many macro series are revised after first release, so the same observation period can have several vintages. Keeping every vintage, instead of overwriting, lets you ask two different questions: what is the newest known value today, and what was the newest value that was actually available on a chosen past date.
What is look-ahead bias?
Look-ahead bias is the error of using information in a historical analysis that was not yet available at the time being analyzed. Using a revised figure for a date before the revision was published, or a monthly statistic before its release day, are common cases. It makes past results look better informed than any real participant could have been.
Why is a three-day-old value current for one series and stale for another?
Freshness is judged against the source's own cadence, not against one shared clock. A weekly report three days after release is waiting for its next scheduled release, so it is current. An intraday feed three hours after its last update has missed several expected updates, so it is stale. A holiday that moves a release date is handled by a holiday list that you supply.
Does a current source mean a trustworthy source?
No. Freshness describes timing and authority describes provenance, and the inspector shows them in separate columns. A current alternative-data series is not more authoritative than a primary regulatory report that is slightly delayed but still within its own expected schedule.
Does this inspector fetch live market data?
No. Live feeds are not connected. The inspector judges only the series records you paste, or a bundled illustrative sample with fictional values and dates. For real values use the primary source, such as the CFTC for Commitments of Traders reports, and paste its publication timestamps into the inspector.
References
The sample data on this page is fictional and drawn from no external dataset. The one primary source below describes the real weekly report used as the model for the weekly example.
- CFTC: Commitments of Traders Explanatory Notes: what the weekly report measures and how its observation and release dates relate.
Jurisdiction: United States. Last reviewed by the Swoopr Editorial Team in October 2026. This page is educational and is not personalized investment advice.
Related Reading
- Sentiment Data Latency, Publication Lags and Vintages: the canonical guide to the dates this tool displays.
- Source Ladder for Market Sentiment: how much weight each kind of source deserves.
- Commitments of Traders: a weekly series observed on Tuesday and published later.
- Margin Debt and Leverage: a monthly series published after month end.